Behind the body: Rashōmon, read by a language model one sentence at a time. Each row is a vector — score, ghost, silence, entropy, pressure, distance, velocity, body — and the ghost columns are the model’s disagreement with itself. This is the live feed.
lil is the line the plotter leaves: the reading, collapsed into ink. [In the prototype the frame opens YouTube; on the site the video plays here.]
Biblio Engine converts Japanese literary texts into nineteen parameters — emotional temperature, narrative velocity, silence density, somatic reaction.
Each sentence is read many times by a language model, and the variance between readings, σ, is kept as the Ghost. Where the Ghost is low, the work holds what the text necessarily is. Where it is high, it holds what reading discovers. At the moment of plotting, the superposition of interpretations collapses into ink.
BIBLIO ENGINEは、日本文学のテキストを19次元のパラメータに変換する。言語モデルによる複数回の読みのばらつき σ を「Ghost」として保持し、プロット（描画）の瞬間に解釈の重ね合わせがインクへと収束する。
A pen plotter draws the collapsed reading onto paper. Ink, once laid, does not change.
The generative token, with its own Ghost sliders, will live on-chain — a version anyone can hold. Platform to be decided.
In the room, the reading is projected and OSC-driven — present only for as long as the room is.